XAI4TSC documentation

XAI4TSC is a Python framework for benchmarking eXplainable AI (XAI) methods on time series classification (TSC) models. It covers the full evaluation pipeline:

  • Data — load UCR/UEA datasets or local files, split, encode, and cache

  • Models — train PyTorch classifiers with a unified ModelBase interface

  • Explanations — generate feature attributions via Captum (Integrated Gradients, DeepLIFT, Occlusion, and more)

  • Evaluation — quantify explanation quality with 38 Quantus metrics

XAI4TSC is designed for two use cases: as an importable package for programmatic use in notebooks and scripts, and as a YAML-driven experiment runner for large-scale reproducible benchmarks.

Contents

Note

This project is under active development.